Latent Space Coding Capsule Network for Mental Workload Classification.

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Real-time mental workload monitoring is crucial for optimizing work efficiency.
• Electroencephalography (EEG) features like band power and brain connectivity aid workload classification.
• Existing deep learning models struggle with non-stationary EEG data, limiting classification accuracy.

Purpose of the Study:

• To propose a novel deep learning model, the latent space coding capsule network (LSCCN), for improved mental workload classification.
• To integrate complementary EEG features (band power and brain connectivity) for robust workload assessment.
• To address the limitations of current models in handling EEG data variability.

Main Methods:

• Developed and applied the latent space coding capsule network (LSCCN) model.
• Fused band power and brain connectivity features within a latent space.
• Utilized convolutional and capsule modules for classification.
• Compared LSCCN performance against state-of-the-art methods.

Main Results:

• LSCCN demonstrated superior performance compared to existing methods.
• Achieved higher testing accuracy and a smaller standard deviation, indicating reliable classification across participants.
• Identified discriminative features localized in frontal, parietal, and occipital regions.

Conclusions:

• The proposed LSCCN model offers a significant advancement in mental workload classification.
• The findings support the integration of diverse EEG features for enhanced monitoring.
• This research paves the way for more practical applications of workload monitoring systems.